diff mbox

[hsa,1/10] Configury changes and new options

Message ID 20151207111908.GB24234@virgil.suse.cz
State New
Headers show

Commit Message

Martin Jambor Dec. 7, 2015, 11:19 a.m. UTC 
Hi,

this patch contains changes to the configuration mechanism and offload
bits, so that users can build compilers with HSA support. It plays
nicely with other accelerators despite using an altogether different
implementation approach.  I have also added to it definitions of the
new options and parameters, since at least one hunk in common.opt is
highly related.  -fdisable-hsa-gridification has disappeared, othrwise
very little has changed since the last submission.

With this patch, the user can request HSA support by including the
string "hsa" among the requested accelerators in
--enable-offload-targets.  This will cause the compiler to start
producing HSAIL for target OpenMP constructs/functions and the hsa
libgomp plugin to be built.  Because the plugin needs to use HSA
run-time library, I have introduced options --with-hsa-runtime (and
more precise --with-hsa-include and --with-hsa-lib) to help find it.
The open-sourced hsa runtime available at github is binary compatible
with the closed-source one which however also contains the finalizer
and so needs to be used for all practical purposes.  I am regularly
asking AMD to keep their promise and open source the finalizer too.

One catch is however that there is no offload compiler for HSA and so
the wrapper should not attempt to look for it (that is what the hunk
in lto-wrapper.c does) and when HSA is the only accelerator, it is
wasteful to output LTO sections with byte-code and therefore if HSA is
the only configured accelerator, it does not set ENABLE_OFFLOADING
macro.

Finally, when the compiler has been configured for HSA but the user
disables it by omitting it in the -foffload compiler option, we need
to observe that decision.  That is what the opts.c hunk does.

As far as the options are concerned, the patch adds new warning -Whsa
we emit whenever we fail to produce HSAIL for some source code.  It is
on by default but warnigs are of course only emitted by HSAIL
generating code so will never affect anybody who does not use both an
HSA-enabled compiler and OpenMP 4 device constructs.

Then there is a new parameter hsa-gen-debug-stores, which will be
obsolete once HSA run-time supports debugging traps.  Before that, we
have to do with debugging stores to memory at defined places, which
however can cost speed in benchmarks.  So we only enabled them with
this parameter.  We decided to make it a parameter rather than a
switch to emphasize the fact it will go away and to possibly allow us
select different levels of verbosity of the stores in the future).

Any feedback is very appreciated,

Martin



2015-12-04  Martin Jambor  <mjambor@suse.cz>

gcc/
	* Makefile.in (OBJS): Add new source files.
	(GTFILES): Add hsa.c.
	* config.in (ENABLE_HSA): New.
	* configure.ac: Treat hsa differently from other accelerators.
	(OFFLOAD_TARGETS): Define ENABLE_OFFLOADING according to
	$enable_offloading.
	(ENABLE_HSA): Define ENABLE_HSA according to $enable_hsa.
	* doc/install.texi (Configuration): Document --with-hsa-runtime,
	--with-hsa-runtime-include and --with-hsa-runtime-lib.
	* lto-wrapper.c (compile_images_for_offload_targets): Do not attempt
	to invoke offload compiler for hsa acclerator.
	* opts.c (common_handle_option): Determine whether HSA offloading
	should be performed.
	* common.opt (disable_hsa): New variable.
	(-Whsa): New warning.
	* doc/invoke.texi (-Whsa): Document.
	(hsa-gen-debug-stores): Likewise.
	* params.def (PARAM_HSA_GEN_DEBUG_STORES): New parameter.

libgomp/plugin/
	* Makefrag.am: Add HSA plugin requirements.
	* configfrag.ac (HSA_RUNTIME_INCLUDE): New variable.
	(HSA_RUNTIME_LIB): Likewise.
	(HSA_RUNTIME_CPPFLAGS): Likewise.
	(HSA_RUNTIME_INCLUDE): New substitution.
	(HSA_RUNTIME_LIB): Likewise.
	(HSA_RUNTIME_LDFLAGS): Likewise.
	(hsa-runtime): New configure option.
	(hsa-runtime-include): Likewise.
	(hsa-runtime-lib): Likewise.
	(PLUGIN_HSA): New substitution variable.
	Fill HSA_RUNTIME_INCLUDE and HSA_RUNTIME_LIB according to the new
	configure options.
	(PLUGIN_HSA_CPPFLAGS): Likewise.
	(PLUGIN_HSA_LDFLAGS): Likewise.
	(PLUGIN_HSA_LIBS): Likewise.
	Check that we have access to HSA run-time.

Comments

Richard Sandiford Dec. 8, 2015, 10:43 p.m. UTC | #1 
[Sorry for the low-quality review, was just reading out of interest...]

Martin Jambor <mjambor@suse.cz> writes:
> +If you configure GCC with HSA offloading but do not have the HSA
> +run-time library installed in a standard location then you can
> +explicitely specify the directory where they are installed.  The

typo: explicitly

> diff --git a/gcc/lto-wrapper.c b/gcc/lto-wrapper.c
> index e4772d1..5609207 100644
> --- a/gcc/lto-wrapper.c
> +++ b/gcc/lto-wrapper.c
> @@ -745,6 +745,11 @@ compile_images_for_offload_targets (unsigned in_argc, char *in_argv[],
>    offload_names = XCNEWVEC (char *, num_targets + 1);
>    for (unsigned i = 0; i < num_targets; i++)
>      {
> +      /* HSA does not use LTO-like streaming and a different compiler, skip
> +	 it. */
> +      if (strncmp(names[i], "hsa", 3) == 0)
> +	continue;
> +
>        offload_names[i]
>  	= compile_offload_image (names[i], compiler_path, in_argc, in_argv,
>  				 compiler_opts, compiler_opt_count,

Looks like this would cause the caller loop:

      if (offload_names)
	{
	  find_offloadbeginend ();
	  for (i = 0; offload_names[i]; i++)
	    printf ("%s\n", offload_names[i]);
	  free_array_of_ptrs ((void **) offload_names, i);
	}

to terminate early if there was another target after hsa.

names[i] is null-terminated, so it looks like you're deliberately
allowing anything that starts with "hsa" here, but:

> diff --git a/gcc/opts.c b/gcc/opts.c
> index 874c84f..5647f0c 100644
> --- a/gcc/opts.c
> +++ b/gcc/opts.c
> @@ -1906,8 +1906,35 @@ common_handle_option (struct gcc_options *opts,
>        break;
>  
>      case OPT_foffload_:
> -      /* Deferred.  */
> -      break;
> +      {
> +	const char *p = arg;
> +	opts->x_flag_disable_hsa = true;
> +	while (*p != 0)
> +	  {
> +	    const char *comma = strchr (p, ',');
> +
> +	    if ((strncmp (p, "disable", 7) == 0)
> +		&& (p[7] == ',' || p[7] == '\0'))
> +	      {
> +		opts->x_flag_disable_hsa = true;
> +		break;
> +	      }
> +
> +	    if ((strncmp (p, "hsa", 3) == 0)
> +		&& (p[3] == ',' || p[3] == '\0'))
> +	      {
> +#ifdef ENABLE_HSA
> +		opts->x_flag_disable_hsa = false;
> +#else
> +		sorry ("HSA has not been enabled during configuration");
> +#endif

...here you only allow "hsa" itself.

(Not your fault, but: do we have any documentation for -foffload
and -foffload-abi?  Couldn't see any in the texi files.)

Thanks,
Richard
diff mbox

Patch

diff --git a/gcc/Makefile.in b/gcc/Makefile.in
index bee2879..5fe73a7 100644
--- a/gcc/Makefile.in
+++ b/gcc/Makefile.in
@@ -1296,6 +1296,11 @@  OBJS = \
 	graphite-sese-to-poly.o \
 	gtype-desc.o \
 	haifa-sched.o \
+	hsa.o \
+	hsa-gen.o \
+	hsa-regalloc.o \
+	hsa-brig.o \
+	hsa-dump.o \
 	hw-doloop.o \
 	hwint.o \
 	ifcvt.o \
@@ -1320,6 +1325,7 @@  OBJS = \
 	ipa-icf.o \
 	ipa-icf-gimple.o \
 	ipa-reference.o \
+	ipa-hsa.o \
 	ipa-ref.o \
 	ipa-utils.o \
 	ipa.o \
@@ -2401,6 +2407,7 @@  GTFILES = $(CPP_ID_DATA_H) $(srcdir)/input.h $(srcdir)/coretypes.h \
   $(srcdir)/sanopt.c \
   $(srcdir)/ipa-devirt.c \
   $(srcdir)/internal-fn.h \
+  $(srcdir)/hsa.c \
   @all_gtfiles@
 
 # Compute the list of GT header files from the corresponding C sources,
diff --git a/gcc/config.in b/gcc/config.in
index bb0d220..eee9f60 100644
--- a/gcc/config.in
+++ b/gcc/config.in
@@ -144,6 +144,12 @@ 
 #endif
 
 
+/* Define this to enable support for generating HSAIL. */
+#ifndef USED_FOR_TARGET
+#undef ENABLE_HSA
+#endif
+
+
 /* Define if gcc should always pass --build-id to linker. */
 #ifndef USED_FOR_TARGET
 #undef ENABLE_LD_BUILDID
diff --git a/gcc/configure.ac b/gcc/configure.ac
index 5990b7c..5ce1881 100644
--- a/gcc/configure.ac
+++ b/gcc/configure.ac
@@ -940,6 +940,13 @@  AC_SUBST(accel_dir_suffix)
 
 for tgt in `echo $enable_offload_targets | sed 's/,/ /g'`; do
   tgt=`echo $tgt | sed 's/=.*//'`
+
+  if echo "$tgt" | grep "^hsa" > /dev/null ; then
+    enable_hsa=1
+  else
+    enable_offloading=1
+  fi
+
   if test x"$offload_targets" = x; then
     offload_targets=$tgt
   else
@@ -948,7 +955,7 @@  for tgt in `echo $enable_offload_targets | sed 's/,/ /g'`; do
 done
 AC_DEFINE_UNQUOTED(OFFLOAD_TARGETS, "$offload_targets",
   [Define to offload targets, separated by commas.])
-if test x"$offload_targets" != x; then
+if test x"$enable_offloading" != x; then
   AC_DEFINE(ENABLE_OFFLOADING, 1,
     [Define this to enable support for offloading.])
 else
@@ -956,6 +963,11 @@  else
     [Define this to enable support for offloading.])
 fi
 
+if test x"$enable_hsa" = x1 ; then
+  AC_DEFINE(ENABLE_HSA, 1,
+    [Define this to enable support for generating HSAIL.])
+fi
+
 AC_ARG_WITH(multilib-list,
 [AS_HELP_STRING([--with-multilib-list], [select multilibs (AArch64, SH and x86-64 only)])],
 :,
diff --git a/gcc/doc/install.texi b/gcc/doc/install.texi
index 0b71bef..232586d 100644
--- a/gcc/doc/install.texi
+++ b/gcc/doc/install.texi
@@ -1982,6 +1982,23 @@  specifying paths @var{path1}, @dots{}, @var{pathN}.
 % @var{srcdir}/configure \
     --enable-offload-target=i686-unknown-linux-gnu=/path/to/i686/compiler,x86_64-pc-linux-gnu
 @end smallexample
+
+If @samp{hsa} is specified as one of the targets, the compiler will be
+built with support for HSA GPU accelerators.  Because the same
+compiler will emit the accelerator code, no path should be specified.
+
+@item --with-hsa-runtime=@var{pathname}
+@itemx --with-hsa-runtime-include=@var{pathname}
+@itemx --with-hsa-runtime-lib=@var{pathname}
+
+If you configure GCC with HSA offloading but do not have the HSA
+run-time library installed in a standard location then you can
+explicitely specify the directory where they are installed.  The
+@option{--with-hsa-runtime=@/@var{hsainstalldir}} option is a
+shorthand for
+@option{--with-hsa-runtime-lib=@/@var{hsainstalldir}/lib} and
+@option{--with-hsa-runtime-include=@/@var{hsainstalldir}/include}.
+
 @end table
 
 @subheading Cross-Compiler-Specific Options
diff --git a/gcc/lto-wrapper.c b/gcc/lto-wrapper.c
index e4772d1..5609207 100644
--- a/gcc/lto-wrapper.c
+++ b/gcc/lto-wrapper.c
@@ -745,6 +745,11 @@  compile_images_for_offload_targets (unsigned in_argc, char *in_argv[],
   offload_names = XCNEWVEC (char *, num_targets + 1);
   for (unsigned i = 0; i < num_targets; i++)
     {
+      /* HSA does not use LTO-like streaming and a different compiler, skip
+	 it. */
+      if (strncmp(names[i], "hsa", 3) == 0)
+	continue;
+
       offload_names[i]
 	= compile_offload_image (names[i], compiler_path, in_argc, in_argv,
 				 compiler_opts, compiler_opt_count,
diff --git a/gcc/opts.c b/gcc/opts.c
index 874c84f..5647f0c 100644
--- a/gcc/opts.c
+++ b/gcc/opts.c
@@ -1906,8 +1906,35 @@  common_handle_option (struct gcc_options *opts,
       break;
 
     case OPT_foffload_:
-      /* Deferred.  */
-      break;
+      {
+	const char *p = arg;
+	opts->x_flag_disable_hsa = true;
+	while (*p != 0)
+	  {
+	    const char *comma = strchr (p, ',');
+
+	    if ((strncmp (p, "disable", 7) == 0)
+		&& (p[7] == ',' || p[7] == '\0'))
+	      {
+		opts->x_flag_disable_hsa = true;
+		break;
+	      }
+
+	    if ((strncmp (p, "hsa", 3) == 0)
+		&& (p[3] == ',' || p[3] == '\0'))
+	      {
+#ifdef ENABLE_HSA
+		opts->x_flag_disable_hsa = false;
+#else
+		sorry ("HSA has not been enabled during configuration");
+#endif
+	      }
+	    if (!comma)
+	      break;
+	    p = comma + 1;
+	  }
+	break;
+      }
 
 #ifndef ACCEL_COMPILER
     case OPT_foffload_abi_:
diff --git a/libgomp/plugin/Makefrag.am b/libgomp/plugin/Makefrag.am
index 745becd..433bba1 100644
--- a/libgomp/plugin/Makefrag.am
+++ b/libgomp/plugin/Makefrag.am
@@ -38,3 +38,16 @@  libgomp_plugin_nvptx_la_LDFLAGS += $(PLUGIN_NVPTX_LDFLAGS)
 libgomp_plugin_nvptx_la_LIBADD = libgomp.la $(PLUGIN_NVPTX_LIBS)
 libgomp_plugin_nvptx_la_LIBTOOLFLAGS = --tag=disable-static
 endif
+
+if PLUGIN_HSA
+# Heterogenous Systems Architecture plugin
+libgomp_plugin_hsa_version_info = -version-info $(libtool_VERSION)
+toolexeclib_LTLIBRARIES += libgomp-plugin-hsa.la
+libgomp_plugin_hsa_la_SOURCES = plugin/plugin-hsa.c
+libgomp_plugin_hsa_la_CPPFLAGS = $(AM_CPPFLAGS) $(PLUGIN_HSA_CPPFLAGS)
+libgomp_plugin_hsa_la_LDFLAGS = $(libgomp_plugin_hsa_version_info) \
+	$(lt_host_flags)
+libgomp_plugin_hsa_la_LDFLAGS += $(PLUGIN_HSA_LDFLAGS)
+libgomp_plugin_hsa_la_LIBADD = libgomp.la $(PLUGIN_HSA_LIBS)
+libgomp_plugin_hsa_la_LIBTOOLFLAGS = --tag=disable-static
+endif
diff --git a/libgomp/plugin/configfrag.ac b/libgomp/plugin/configfrag.ac
index ad70dd1..c50e5cb 100644
--- a/libgomp/plugin/configfrag.ac
+++ b/libgomp/plugin/configfrag.ac
@@ -81,6 +81,54 @@  AC_SUBST(PLUGIN_NVPTX_CPPFLAGS)
 AC_SUBST(PLUGIN_NVPTX_LDFLAGS)
 AC_SUBST(PLUGIN_NVPTX_LIBS)
 
+# Look for HSA run-time, its includes and libraries
+
+HSA_RUNTIME_INCLUDE=
+HSA_RUNTIME_LIB=
+AC_SUBST(HSA_RUNTIME_INCLUDE)
+AC_SUBST(HSA_RUNTIME_LIB)
+HSA_RUNTIME_CPPFLAGS=
+HSA_RUNTIME_LDFLAGS=
+
+AC_ARG_WITH(hsa-runtime,
+	[AS_HELP_STRING([--with-hsa-runtime=PATH],
+		[specify prefix directory for installed HSA run-time package.
+		 Equivalent to --with-hsa-runtime-include=PATH/include
+		 plus --with-hsa-runtime-lib=PATH/lib])])
+AC_ARG_WITH(hsa-runtime-include,
+	[AS_HELP_STRING([--with-hsa-runtime-include=PATH],
+		[specify directory for installed HSA run-time include files])])
+AC_ARG_WITH(hsa-runtime-lib,
+	[AS_HELP_STRING([--with-hsa-runtime-lib=PATH],
+		[specify directory for the installed HSA run-time library])])
+if test "x$with_hsa_runtime" != x; then
+  HSA_RUNTIME_INCLUDE=$with_hsa_runtime/include
+  HSA_RUNTIME_LIB=$with_hsa_runtime/lib
+fi
+if test "x$with_hsa_runtime_include" != x; then
+  HSA_RUNTIME_INCLUDE=$with_hsa_runtime_include
+fi
+if test "x$with_hsa_runtime_lib" != x; then
+  HSA_RUNTIME_LIB=$with_hsa_runtime_lib
+fi
+if test "x$HSA_RUNTIME_INCLUDE" != x; then
+  HSA_RUNTIME_CPPFLAGS=-I$HSA_RUNTIME_INCLUDE
+fi
+if test "x$HSA_RUNTIME_LIB" != x; then
+  HSA_RUNTIME_LDFLAGS=-L$HSA_RUNTIME_LIB
+fi
+
+PLUGIN_HSA=0
+PLUGIN_HSA_CPPFLAGS=
+PLUGIN_HSA_LDFLAGS=
+PLUGIN_HSA_LIBS=
+AC_SUBST(PLUGIN_HSA)
+AC_SUBST(PLUGIN_HSA_CPPFLAGS)
+AC_SUBST(PLUGIN_HSA_LDFLAGS)
+AC_SUBST(PLUGIN_HSA_LIBS)
+
+
+
 # Get offload targets and path to install tree of offloading compiler.
 offload_additional_options=
 offload_additional_lib_paths=
@@ -122,6 +170,49 @@  if test x"$enable_offload_targets" != x; then
 	    ;;
 	esac
 	;;
+      hsa*)
+	case "${target}" in
+	  x86_64-*-*)
+	    case " ${CC} ${CFLAGS} " in
+	      *" -m32 "*)
+	        PLUGIN_HSA=0
+		;;
+	      *)
+	        tgt_name=hsa
+	        PLUGIN_HSA=$tgt
+	        PLUGIN_HSA_CPPFLAGS=$HSA_RUNTIME_CPPFLAGS
+	        PLUGIN_HSA_LDFLAGS=$HSA_RUNTIME_LDFLAGS
+	        PLUGIN_HSA_LIBS="-lhsa-runtime64 -lhsakmt"
+
+	        PLUGIN_HSA_save_CPPFLAGS=$CPPFLAGS
+	        CPPFLAGS="$PLUGIN_HSA_CPPFLAGS $CPPFLAGS"
+	        PLUGIN_HSA_save_LDFLAGS=$LDFLAGS
+	        LDFLAGS="$PLUGIN_HSA_LDFLAGS $LDFLAGS"
+	        PLUGIN_HSA_save_LIBS=$LIBS
+	        LIBS="$PLUGIN_HSA_LIBS $LIBS"
+
+	        AC_LINK_IFELSE(
+	          [AC_LANG_PROGRAM(
+	            [#include "hsa.h"],
+	              [hsa_status_t status = hsa_init ()])],
+	          [PLUGIN_HSA=1])
+	        CPPFLAGS=$PLUGIN_HSA_save_CPPFLAGS
+	        LDFLAGS=$PLUGIN_HSA_save_LDFLAGS
+	        LIBS=$PLUGIN_HSA_save_LIBS
+	        case $PLUGIN_HSA in
+	          hsa*)
+	            HSA_PLUGIN=0
+	            AC_MSG_ERROR([HSA run-time package required for HSA support])
+	            ;;
+	        esac
+		;;
+	      esac
+    	    ;;
+	  *-*-*)
+	    PLUGIN_HSA=0
+            ;;
+        esac
+        ;;
       *)
 	AC_MSG_ERROR([unknown offload target specified])
 	;;
@@ -145,3 +236,6 @@  AC_DEFINE_UNQUOTED(OFFLOAD_TARGETS, "$offload_targets",
 AM_CONDITIONAL([PLUGIN_NVPTX], [test $PLUGIN_NVPTX = 1])
 AC_DEFINE_UNQUOTED([PLUGIN_NVPTX], [$PLUGIN_NVPTX],
   [Define to 1 if the NVIDIA plugin is built, 0 if not.])
+AM_CONDITIONAL([PLUGIN_HSA], [test $PLUGIN_HSA = 1])
+AC_DEFINE_UNQUOTED([PLUGIN_HSA], [$PLUGIN_HSA],
+  [Define to 1 if the HSA plugin is built, 0 if not.])
diff --git a/gcc/common.opt b/gcc/common.opt
index e1617c4..1669441 100644
--- a/gcc/common.opt
+++ b/gcc/common.opt
@@ -229,6 +229,10 @@  unsigned int flag_sanitize_recover = SANITIZE_UNDEFINED | SANITIZE_NONDEFAULT |
 Variable
 bool dump_base_name_prefixed = false
 
+; Flag whether HSA generation has been explicitely disabled
+Variable
+bool flag_disable_hsa = false
+
 ###
 Driver
 
@@ -583,6 +587,10 @@  Wfree-nonheap-object
 Common Var(warn_free_nonheap_object) Init(1) Warning
 Warn when attempting to free a non-heap object.
 
+Whsa
+Common Var(warn_hsa) Init(1) Warning
+Warn when a function cannot be expanded to HSAIL.
+
 Winline
 Common Var(warn_inline) Warning
 Warn when an inlined function cannot be inlined.
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index 586f11f..37a07bc 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -295,7 +295,7 @@  Objective-C and Objective-C++ Dialects}.
 -Wunused-but-set-parameter -Wunused-but-set-variable @gol
 -Wuseless-cast -Wvariadic-macros -Wvector-operation-performance @gol
 -Wvla -Wvolatile-register-var  -Wwrite-strings @gol
--Wzero-as-null-pointer-constant}
+-Wzero-as-null-pointer-constant -Whsa}
 
 @item C and Objective-C-only Warning Options
 @gccoptlist{-Wbad-function-cast  -Wmissing-declarations @gol
@@ -5685,6 +5685,10 @@  Suppress warnings when a positional initializer is used to initialize
 a structure that has been marked with the @code{designated_init}
 attribute.
 
+@item -Whsa
+Issue a warning when HSAIL cannot be emitted for the compiled function or
+OpenMP construct.
+
 @end table
 
 @node Debugging Options
@@ -11210,6 +11214,12 @@  dynamic, guided, auto, runtime).  The default is static.
 Maximum depth of recursion when querying properties of SSA names in things
 like fold routines.  One level of recursion corresponds to following a
 use-def chain.
+
+@item hsa-gen-debug-stores
+Enable emission of special debug stores within HSA kernels which are
+then read and reported by libgomp plugin.  Generation of these stores
+is disabled by default, use @option{--param hsa-gen-debug-stores=1} to
+enable it.
 @end table
 @end table
 
diff --git a/gcc/params.def b/gcc/params.def
index 41fd8a8..411d72c 100644
--- a/gcc/params.def
+++ b/gcc/params.def
@@ -1177,6 +1177,11 @@  DEFPARAM (PARAM_MAX_SSA_NAME_QUERY_DEPTH,
 	  "Maximum recursion depth allowed when querying a property of an"
 	  " SSA name.",
 	  2, 1, 0)
+
+DEFPARAM (PARAM_HSA_GEN_DEBUG_STORES,
+	  "hsa-gen-debug-stores",
+	  "Level of hsa debug stores verbosity",
+	  0, 0, 1)
 /*
 
 Local variables: